Conservation laws by symmetries and adjoint symmetries

Conservation laws are formulated for systems of differential equations by using symmetries and adjoint symmetries, and an application to systems of evolution equations is made, together with illustrative examples. The formulation does not require the existence of a Lagrangian for a given system, and the presented examples include computations of conserved densities for the heat equation, Burgers' equation and the Korteweg-de Vries equation.


Introduction
For systems of Euler-Lagrange equations, Noether's theorem shows that symmetries lead to conservation laws [1,2]. The Lagrangian formulation of the systems under consideration is essential in presenting conservation laws from symmetries, and many physically important examples can be found in [1]- [4]. Is it possible to extend such a connection between conservation laws and symmetries for systems of non-Euler-Lagrange equations? A definite answer has been given in the case of systems of discrete evolution equations [5]. We would, in this paper, like to formulate a similar theory of conservation laws for systems of differential equations and apply it to systems of evolution equations. More precisely, we want to exhibit that pairs of symmetries and adjoint symmetries lead to conservation laws for whatever systems of differential equations.
For totally nondegenerate systems of differential equations, conservation laws are classified by characteristic forms [2]. The characteristics, also called the multipliers [6], of conservation laws are adjoint symmetries, and thus, the existence of an adjoint symmetry is necessary for a totally nondegenerate system of differential equations to admit a conservation law. Adjoint symmetries generate conservation laws, when the variational derivatives of product functionals of the adjoint symmetries and the systems under consideration vanish [6]. For systems of evolution equations, functionals are conserved if and only if their variational derivatives are adjoint symmetries [5,7]. Nonlinear self-adjointness has also been introduced on the basis of adjoint systems to construct conservation laws for nonlinearly self-adjoint systems of differential equations, where the nonlinear self-adjointness means that the second set of dependent variables in an adjoint system stands for an adjoint symmetry [8,9]. We will directly utilize both symmetries and adjoint symmetries to produce conservation laws for systems of differential equations, without using any Lagrangian or extended Lagrangian. This paper is organized as follows. In section 2, a formulation of conservation laws is furnished for systems of differential equations, regardless of the existence of a Lagrangian. In section 3, an application is made for systems of evolution equations, and in section 4, the resulting theory is used to compute conserved densities for the heat equation, Burgers' equation and the Korteweg-de Vries equation, along with many new conserved densities. Finally in section 5, a few of concluding remarks are given with some discussion.

A formulation of conservation laws
We will use a plain language to formulate the correspondence between conservation laws and pairs of symmetries and adjoint symmetries, although there is a geometrical language.
for α = (α 1 , · · · , α p ) with non-negative integers α i , 1 ≤ i ≤ p. Assume that A denotes the space of all functions f (x, u, · · · , u (n) ) with n ≥ 0, where f is a smooth function of the involved variables and u (n) is the set of n-th order partial derivatives of u with respect to x, and B denotes the space of all smooth functions of x, u and derivatives of u with respect to x to some finite order. The locality means that a function f (u) depends locally on u with respect to x, i.e., any value (f (u))(x) is completely determined by the value of u in a sufficiently small region of x. Any functions in A, particularly differential polynomial functions, are local. The space B contains nonlocal functions, and simple examples are x l x k 0 ∂ j u i dx k , j = k. We use A r and B r to denote the r-th order tensor products of A and B: Two functions f 1 , f 2 ∈ B are said to be equivalent and denoted by f 1 ∼ f 2 , if there exist g i ∈ B, 1 ≤ i ≤ p, such that This is an equavalence relation. Each equivalence class is called a functional and the class that P ∈ B belongs to is denoted by P dx. The inner products are defined by and the adjoint operator Φ * : B s → B r of a linear operator Φ : B r → B s is determined by For any vector function X = X(u) = (X 1 , · · · , X r ) T ∈ A r , we introduce its Gateaux operator Now, for a given integer l ≥ 1, let us consider a system of differential equations ∆(u) = 0, ∆ = (∆ 1 , · · · , ∆ l ) T ∈ A l . (2.6) Its linearized system and adjoint linearized system are defined by respectively. Here ∆ ′ and (∆ ′ ) * denote the Gateaux operator of ∆ and its adjoint operator, respectively. It is easy to observe that Definition 2.1 A vector function σ ∈ B q is called a symmetry of (2.6), if it satisfies (2.7) when u solves (2.6). A vector function ρ ∈ B l is called an adjoint symmetry of (2.6), if it satisfies (2.8) when u solves (2.6).

9)
holds for all solutions of (2.6), then (2.9) is called a conservation law and P a conserved vector of (2.6).
Conservation laws include familiar concepts of conservation of mass, energy and momentum arising in physical applications. In the case of Laplace's equation, the equation itself is a conservation law, since ∆u = Div (∇u) = 0 for all solutions. For a system of ordinary differential equations involving a single independent variable x ∈ R, a conservation law D x P = 0 requires that P (x, u (u) ) be constant for all solutions of the system. Thus, a conservation law for a system of ordinary differential equations is equivalent to the classical notion of a first integral or constant of motion of the system.
In a dynamical problem, one of the independent variables is distinguished as the time t, the remaining variables x = (x 1 , · · · , x p ) being spatial variables. Then, a conservation law takes the form D t T + Div X = 0, in which Div stands for the spatial divergence of X ∈ B p with respect to x. Here T is called a conserved density and X, the associated flux vector corresponding to T .
Conservation laws of a totally nondegenerate system of differential equations (2.6) are classified through characteristic forms [2]: where Q ∈ B l . Such a vector function Q is called the characteristic of the associated conservation law. For a normal, totally nondegenerate system of differential equations, equivalent conservation laws correspond to equivalent characteristics [2].
We would like to formulate conservation laws for a general system of differential equations (2.6) by using pairs of symmetries and adjoint symmetries. In our expressions of construction, we will use the following assumption for brevity: an empty product of derivatives ∂ α i i , α i ≥ 0, 1 ≤ i ≤ p, is understood to be the identity operator. For example, Π s i=r ∂ 2 i implies the identity operator, when r > s. Lemma 2.1 Let f and g be two smooth functions in variables x 1 , · · · , x p . Then for any where an empty sum is understood to be zero.
Proof: First note that we have for any two smooth functions a and b in variables x 1 , · · · , x p , and then we decompose that It follows from (2.12) that each term in the above sum can be computed as follows: where an empty sum is understood to be zero. This allows us to conclude that our equality (2.11) holds for any α = (α 1 , · · · , α p ). The proof is finished.
This lemma tells us that f D α g − ((−D) α f )g is a total divergence function for any two smooth functions f and g, and also guarantees that (2.5) presents an adjoint operator of X ′ . Theorem 2.1 Let σ = (σ 1 , · · · , σ q ) T ∈ B q and ρ = (ρ 1 , · · · , ρ l ) T ∈ B l be a symmetry and an adjoint symmetry of a system of differential equations (2.6), respectively. Then we have a conservation law for the system (2.6): (2.13) where α = (α 1 , · · · , α p ), and an empty sum is understood to be zero.
Proof: Let us compute that By using Lemma 2.1, for all 1 ≤ i ≤ l, 1 ≤ j ≤ q and α ≥ 0, we have where an empty sum is understood to be zero. Now, noting that ∆ ′ σ = 0 and (∆ ′ ) * ρ = 0 hold for all solutions of (2.6), we see that (2.13) follows. The proof is finished.
The theorem gives us an explicit formulation of conservation laws for systems of differential equations, regardless of the existence of a Lagrangian. For totally nondegenerate systems of differential equations, we can have where R sym σ and R asym ρ are two l × l and q × l matrix differential operators depending on σ and ρ, respectively [2], and then the following computation presents the characteristic of an equivalent conservation law: A direct way to generate more conservation laws from known ones is to use recursion structures. Similar to the definition of recursion operators [10], or hereditary symmetry operators [11], we can also have recursion structures of other kinds for systems of differential equations. Recursion operators transforms symmetries to symmetries, and hereditary symmetry operators provide recursion operators for hierarchies.
All the above operators are important in establishing more conservation laws, and thus integrability of systems of differential equations. The concepts of Noether operators and inverse Noether operators were also introduced for integrable systems [12].
3 Application to evolution equations 3

.1 Conservation laws
Let us take a set of independent variables (t, x 1 , · · · , x p ), including a distinguished time variable t ∈ R, and consider a system of evolution equations (3.1) Obvisouly, we have ∆ = u t − K(u) with l = q, and thus its linearized system and adjoint linearized system read respectively. Here K ′ and (K ′ ) * stand for the Gateaux operator of K and its adjoint operator, respectively.
It is easy to see that two vector functions σ, ρ ∈ A q are a symmetry and an adjoint symmetry of the system (3.1) if and only if they satisfy respectively, when u solves (3.1), where σ ′ and ρ ′ are the Gateaux operators of σ and ρ.
A total divergence gives us a conservation law for the system of evolution equations (3.1), and T is a conserved density of (3.1) and X = (X 1 , · · · , X p ) T , a conversed flux vector of (3.1) corresponding to T .
An application of Theorem 2.1 to systems of evolution equations presents the following result on conservation laws for systems of evolution equations.
Theorem 3.1 Let σ = (σ 1 , · · · , σ q ) T ∈ B q and ρ = (ρ 1 , · · · , ρ q ) T ∈ B q be a symmetry and an adjoint symmetry of a system of evolution equations (3.1), respectively. Then we have a dynamical conservation law for the system (3.1): where the conserved fluxes are defined by where 1 ≤ k ≤ p, α = (α 1 , · · · , α p ), and an empty sum is understood to be zero. Therefore, T = σ T ρ is a conserved density of the system (3.1), and X = (X 1 , · · · , X p ) T ∈ B p is the conserved flux vector of the system (3.1), corresponding to T .
Proof: A simple application of Theorem 2.1 to the case of ∆ = u t − K(u) with a set of independent variables (t, x 1 , · · · , x p ) presents the result in the above thorem.
Alternatively, we can directly prove the theorem. First, we can have Then, together with Lemma 2.1 for all 1 ≤ i, j ≤ q and α ≥ 0, we can see that (3.7) holds for all solutons of (3.1) and the conserved flux vector X is given by (3.8). The proof is finished.
The theorem gives us a direct formulation of conservation laws for systems of evolution equations, and all expressions are explicitly given for the conserved density and conserved fluxes. The involved conserved density is just a product of a symmetry and an adjoint symmetry, but the conserved fluxes are dependent on the pair of a symmetry and an adjoint symmetry and the system itself.

Recursion structures
For systems of evolution equations, we can easily prove the following theorem, which states sufficient conditions for being recursion operators, adjoint recursion operators, Noether operators or inverse Noether operators.
respectively, where K ′ and (K ′ ) * denote the Gateaux operator of K and its adjoint operator, and the Gateaux operator of an operator A is similarly defined by When the operators Φ, Ψ,Φ,Ψ don't explicitly depend on t, the above theorem presents the results established in [12]. It is also easy to see that we have two relations Φ * = Ψ,Φ −1 =Ψ, (3.14) which means that the adjoint operator of a recursion operator (or an adjoint recursion operator) of (3.1) is an adjoint recursion operator (or a recursion operator) of the same system, and that the inverse operator of a Noether operator (or an inverse Noether operator) of (3.1) is an inverse Noether operator (or a Noether operator) of the same system. It is also easy to prove that the inverse operator Φ −1 (or Ψ −1 ) of a recursion operator Φ (or an adjoint recursion operator Ψ) of a system of evolution equations is still a recursion operator (or an adjoint recursion operator) of the same system, if it exists.
Definition 3.1 If a conserved density T ∈ B of a system of evolution equations (3.1) is equivalent to zero function, then T is called trivial, and otherwise T is called nontrivial.
Any spatial divergence function p i=1 ∂ i f i must be a trivial conserved density of any system of evolution equations. According to the definition (2.3) of adjoint operators, we can have (3.15) where Ψ = Φ * . The study of conservation laws aims at presenting nontrivial conserved densities. Thus, we only need to consider σ T (Ψ i ρ), i, j ≥ 0, in order to generate nontrivial and nonequivalent conserved densities, while using recursion operators.
For the Korteweg-de Vries (KdV) equation it is easy to get and then a recursion operator, an adjoint recursion operator, a Noether operator, and an inverse Noether operator of the KdV equation (3.16): Therefore, the KdV equation (3.16) possesses infinitely many symmetries and adjoint symmetries, Φ i σ and Ψ i ρ, i ≥ 0, once we have a pair of a symmetry σ and an adjoint symmetry ρ. We will see later that this really happens.

Illustrative examples
Now we go on to illustrate by examples rich structures of the conserved densities resulting from symmetries and adjoint symmetries.

The heat equation
Let us consider the heat equation which lacks a Lagrangian formulation. Its linearized equation and adjoint linearized equation read as respectively.
Among the first-order differential operators, it is easy to obtain two recursion operators of (4.1): and thus two adjoint recursion operators of (4.1): Obviously, any solution f = f (x, t) to f t = f xx and the function K 0 = u are symmetries of the heat equation (4.1), and any solution g = g(x, t) to g t = −g xx and the function S 0 = u(x, −t) are adjoint symmetries of the heat equation (4.1). Here the adjoint symmetry S 0 has a local dependence on u with respect to x, similar to the symmetry K 0 . Only for linear equations, adjoint symmetries can have such a local dependence on u.
Now, by the principle in Theorem 3.1, we have infinitely many conserved densities besides a class of trivial conserved densities given by f g. Noting that we have (3.15) and we did not list the other equivalent or linear conbination type conserved densities such as whereḡ is a new solution to g t = −g xx . The first class of conserved densities gu in (4.6) presents the ones generated in [6], and the third class of conserved densities contains the following interesting conserved densities:

Burgers' equation
Let us now consider Burgers' equation which lacks a Lagrangian formulation as well. Its linearized equation and adjoint linearized equation readσ respectively.
Since the Cole-Hope transformation linearizes Burgers' equation (4.10) to the heat equation (4.1) with a dependent variable v, we can move all above results for the heat equation to Burgers' equation. Note that the Gateaux operator of B and its inverse operator read (4.14) The Cole-Hope transformation gives us two recursion operators for Burgers' equation (4.10): and two relations on symmetries and adjoint symmetries between two equations Now, the conserved densities by the principle in Theorem 3.1 can be computed as follows: Therefore, for example, if we choose σ = g and ρ = v, where g solves g t = −g xx , then we can immediately obtain a class of conserved densities for Burgers' equation (4.10): which was generated in [6]. In fact, this class of conserved densities corresponds to the following conservation laws Let us set two basic symmetries of Burgers' equation (4.10): 20) where f solves f t = f xx . Since there exist two inverse recursion operators [13]: (4.22) where the inverse in the middle of the second formula can be worked out: we can have infinitely many symmetries for Burgers' equation (4.10): where we can not add the other symmetries such asΦ 1Φ2Φ1K0 to the algebra spanned by all symmetries in (4.23) because The symmetries defined by (4.23) contain all symmetries generated in [14], and in particular, we have a time-dependent symmetrỹ Moreover, a direct computation can show that all local adjoint symmetries of (4.10), depending on x, t, u, and derivatives of u with respect to x to some finite order, must be a constant function. By noting that x , all conserved densities resulted from the products of the above symmetries and a local adjoint symmetryρ 0 = 1 must be trivial. For example, the following class of conserved densities is trivial. This result also provides an evidence why Burgers' equation (4.10) has only one nontrivial conserved density of differential polynomial type [15].
Nevertheless, based on two basic adjoint symmetries of the heat equation, we can obtain two nonlocal basic adjoint symmetries of Burgers' equation (4.10), defined bỹ where g solves g t = −g xx and S 0 (x, t) = u(x, −t). Now, by the principle in Theorem 3.1, we have infinitely many conserved densities Several simple classes of conserved densities can be computed as follows: They correspond to the following conservation laws the first and the fourth of which are trivial conservation laws of the second kind (see [2] for the definition). Of course, there are the other two classes of conserved densities, generated from the second basic adjoint symmetryS 0 .

The Korteweg-de Vries equation
Let us finally consider the Korteweg-de Vries (KdV) equation (3.16), i.e., This standard form of the KdV equation also lacks a Lagrangian formulation. Its linearized equation and adjoint linearized equation read respectively. It is known that we have infinitely many symmetries with Φ being given by (3.18). They form a Virasoro algebra (see, e.g., [16]): where K −1 = τ −1 = 0, and the commutator is defined by In particular, we have but the other time-dependent symmetries τ i , i ≥ 2, are all nonlocal.
Since we have an inverse Noether operator ∂ −1 x , we can immediately obtain two local adjoint symmetries of the KdV equation (3.16): where P i and Q i are N dimensional vector functions, Therefore, we can have infinitely many conserved densities in terms of eigenfunctions and adjoint eigenfunctions: the first two conserved densities of which are the following φ 1s ψ 2s .

Conclusion and remarks
We have established a direct formulation of conservation laws for systems of differential equations, regardless of the existence of a Lagrangian, and made an application to systems We remark that pairs of symmetries and adjoint symmetries have also been used to formulate conservation laws for systems of discrete evolution equations [5], and Theorem 2.1 could be obtained from an application of Noether's theorem to an enlarged Euler-Lagrange system δL δv = 0, δL δu = 0, (5.1) where L = v T ∆(u) dx with v = (v 1 , · · · , v l ) T . The first subsystem is precisely the original system ∆ = 0, and the second subsystem δL δu = (∆ ′ ) * v = 0 is satisfied if we take v as an adjoint symmetry ρ(u) of the system ∆ = 0. When v = ρ = u (or v = ρ(x, u)), the system under consideration is called strictly (or nonlinearly) self-adjoint, and the resulting conservation law presents the one in [8,9]. In the case of self-adjoint systems of differential equations, Theorem 2.1 generates trivial conservation laws of the second kind, since symmetries are adjoint symmetries, too [18]. The idea of constructing conserved densities by symmetries and adjoint symmetries is also similar to that of binary symmetry constraints, which results in a binary nonlinearization theory [19].
There exists a differential geometric formulation for attempting adjoint symmetries of the second-order ordinary differential equations [20], and adjoint symmetries are also used to show separability of Hamiltonian systems of ordinary differential equations [21]. The existence of symmetry algebras are due to Lie algebraic structures associated with Lax operators corresponding to symmetries (see [22,23,24] for systems of continuous evolution equations and [25,26] for systems of discrete evolution equations). How about Lie algebraic structures for adjoint symmetries? What kind of commutators between adjoint symmetries can be introduced? A good candidate for commutators could be taken as where (ρ ′ 1 ) * and (ρ ′ 2 ) * denote the adjoint operator of their Gateaux operators. However, this doesn't keep the space of adjoint symmetries closed. This can be seen from an example in the case of the KdV equation. The KdV equation (3.16) has two adjoint symmetries ρ 1 = u and ρ 2 = 3 2 tu + x, whose expected commutator reads But the resulting function x is not an adjoint symmetry of the KdV equation (3.16). The characteristic in (2.16) may be useful in exploring a successful commutator of adjoint symmetries.